Why So Many AI Initiatives Are Failing (And What It Means for Insurance)
By Nick Johnson, Director of Technology
A long-time customer of one of our member agencies recently tried to file a claim. She has been with the same agency for years, knows the staff by name, and has never had a serious problem. This time, the carrier had deployed an AI-powered claims intake system. She spent far too much time going in circles, unable to get the system to understand her situation, and when she tried to reach a human being, that path had been deliberately buried. She eventually gave up and called the agency directly, frustrated and shaken.
This story is not unique. Versions of it are playing out across insurance, healthcare, banking, and retail every day. AI is being deployed at a pace that far outstrips the ability to deploy it well, and the damage is not just operational. It falls on the customer relationships that businesses depend on.
The Rush to Deploy
The pressure to adopt AI is enormous right now. Boards are asking about it. Competitors are announcing it. Vendors are selling it. The incentive is to move fast and show something, and that pressure produces a predictable failure mode: companies deploy AI in customer-facing roles before honestly answering the most important question, which is what problem they are actually solving and whether AI is the right tool for it.
Claims intake is a revealing example. The efficiency case is real: fewer staff hours, faster routing, lower cost per interaction. But claims intake is not a data entry problem. It is a human moment. A person filing a claim has just experienced something bad and the interaction that follows is the first signal about whether the relationship with the insurer will hold up under pressure. Optimizing that moment purely for efficiency is a category error.
AI is genuinely good at processing structured data, finding patterns at scale, and performing repeatable tasks without fatigue. It is not good at reading emotional context, handling ambiguity, or knowing when to hand off to a person. Deploying it in roles that require those things, and then making it hard to reach a human, is not an AI problem. It is a design problem.
The Gap Between Demos and Reality
AI systems tend to perform well in controlled demonstrations and struggle in the real world. Real users do not phrase things the way the training data expected. They interrupt, provide context out of order, and describe the same situation ten different ways. When AI handles this poorly, it does not fail gracefully. It loops, misclassifies the issue, and keeps going confidently in the wrong direction.
This is compounded by underinvestment in what happens after launch. AI projects get substantial resources for development and deployment, and far fewer for the ongoing monitoring that would catch failures in production. The system goes live, the project is declared a success, and the people who built it move on. The customers who are struggling have nowhere to go.
The Human Escape Hatch Problem
The single most consequential design decision in any customer-facing AI system is how easy it is to reach a human being. Many organizations have made it hard, burying that path or limiting it to certain hours, on the logic that if the escape is too easy, everyone will use it and the efficiency gains disappear.
The customer logic runs the other way. If I cannot get help when the AI fails me, the experience is worse than having no AI at all. Research bears this out: customers who have a bad automated experience and cannot reach a human are significantly more likely to leave and to tell others about it. The efficiency savings get eaten by attrition and reputation damage.
For insurance, the stakes are higher than in most industries. Insurance relationships are built on the expectation that when something goes wrong, someone will be there. An AI system that blocks access to human help at exactly that moment is not a feature. It is a liability, and increasingly a regulatory one. The Department of Financial Services has been paying closer attention to algorithmic accountability in insurance contexts, and organizations that have deployed AI without clear human escalation paths are building exposure they may not have fully accounted for.
What Good Looks Like
None of this is an argument against AI. It is an argument for deploying it in the right places, with the right safeguards.
Good implementations start with an honest assessment of where automation actually helps the customer, not just the business. In insurance, that includes document processing, coverage verification, policy lookups, and claims status updates where someone has a simple question and wants a fast answer. It is less suited to first-notice-of-loss intake, coverage disputes, or any interaction where emotional context matters.
Good implementations make the path to a human obvious, and they monitor how often people use it. A high escalation rate is a signal that the AI is not handling what it was designed for. It is not a reason to hide the door.
And good implementations define clear ownership. Someone has ongoing responsibility for how the system performs with real customers, with the authority to intervene when it falls short. Without that, there is no accountability when things go wrong, and things will go wrong.
What This Means for Independent Agents
Independent agents can do something carriers and vendors cannot: have a real conversation with clients about their experience. When a client comes to you after a frustrating encounter with an automated system, that is information worth documenting and escalating when possible. That ground-level signal is not reaching the people who built these systems, and it should be.
The same questions apply when deploying AI in your own agency. Before adding any AI-powered tool to a customer-facing workflow, ask what problem it solves, how a customer gets help when it does not work, and who in the agency owns the outcome. Those questions are not always easy to answer. They are always worth asking.
The agencies that will get the most out of AI are not the ones who move fastest. They are the ones who move carefully, with a clear-eyed view of what the technology can and cannot do, and with the human relationships intact that have always been the foundation of this business.
At the end of her rope with an AI that could not help her, our member’s customer called her agent. That is not a failure of technology. That is a reminder of what independent agents are for.
Topics
